tools/lib/perf/evlist.c | 9 +++ tools/lib/perf/include/internal/evlist.h | 2 + tools/perf/Documentation/perf-record.txt | 3 + tools/perf/builtin-record.c | 92 +++++++++++++++------- tools/perf/tests/attr/system-wide-dummy | 14 ++-- tools/perf/tests/attr/test-record-C0 | 4 +- tools/perf/tests/attr/test-record-dummy-C0 | 55 +++++++++++++ tools/perf/tests/shell/record_sideband.sh | 58 ++++++++++++++ tools/perf/util/evlist.c | 18 +++++ tools/perf/util/evlist.h | 1 + 10 files changed, 221 insertions(+), 35 deletions(-) create mode 100644 tools/perf/tests/attr/test-record-dummy-C0 create mode 100755 tools/perf/tests/shell/record_sideband.sh
User space tasks can migrate between CPUs, track sideband events for all
CPUs.
The specific scenarios are as follows:
CPU0 CPU1
perf record -C 0 start
taskA starts to be created and executed
-> PERF_RECORD_COMM and PERF_RECORD_MMAP
events only deliver to CPU1
......
|
migrate to CPU0
|
Running on CPU0 <----------/
...
perf record -C 0 stop
Now perf samples the PC of taskA. However, perf does not record the
PERF_RECORD_COMM and PERF_RECORD_COMM events of taskA.
Therefore, the comm and symbols of taskA cannot be parsed.
The sys_perf_event_open invoked is as follows:
# perf --debug verbose=3 record -e cpu-clock -C 1 true
<SNIP>
Opening: cpu-clock
------------------------------------------------------------
perf_event_attr:
type 1 (PERF_TYPE_SOFTWARE)
size 136
config 0 (PERF_COUNT_SW_CPU_CLOCK)
{ sample_period, sample_freq } 4000
sample_type IP|TID|TIME|CPU|PERIOD|IDENTIFIER
read_format ID|LOST
disabled 1
inherit 1
freq 1
sample_id_all 1
exclude_guest 1
------------------------------------------------------------
sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0x8 = 5
Opening: dummy:u
------------------------------------------------------------
perf_event_attr:
type 1 (PERF_TYPE_SOFTWARE)
size 136
config 0x9 (PERF_COUNT_SW_DUMMY)
{ sample_period, sample_freq } 1
sample_type IP|TID|TIME|CPU|IDENTIFIER
read_format ID|LOST
inherit 1
exclude_kernel 1
exclude_hv 1
mmap 1
comm 1
task 1
sample_id_all 1
exclude_guest 1
mmap2 1
comm_exec 1
ksymbol 1
bpf_event 1
------------------------------------------------------------
sys_perf_event_open: pid -1 cpu 0 group_fd -1 flags 0x8 = 6
sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0x8 = 7
sys_perf_event_open: pid -1 cpu 2 group_fd -1 flags 0x8 = 9
sys_perf_event_open: pid -1 cpu 3 group_fd -1 flags 0x8 = 10
sys_perf_event_open: pid -1 cpu 4 group_fd -1 flags 0x8 = 11
sys_perf_event_open: pid -1 cpu 5 group_fd -1 flags 0x8 = 12
sys_perf_event_open: pid -1 cpu 6 group_fd -1 flags 0x8 = 13
sys_perf_event_open: pid -1 cpu 7 group_fd -1 flags 0x8 = 14
<SNIP>
Changes since_v7:
- The condition for requiring system_wide sideband is changed to
"as long as a non-dummy event exists" (patch4).
- Modify the corresponding test case to record only dummy event (patch6).
- Thanks to tested-by tag from Ravi, but because the solution is modified,
the tested-by tag of Ravi is not added to this version.
Changes since_v6:
- Patch1:
1. No change.
2. Keep Acked-by tag from Adrian.
- Patch2:
1. Update commit message as suggested by Ian.
2. Keep Acked-by tag from Adrian because code is not modified.
- Patch3:
1. Update comment as suggested by Ian.
2. Merge original patch5 ("perf test: Update base-record & system-wide-dummy attr") as suggested by Ian.
3. Only merge commit, keep Acked-by tag from Adrian.
- Patch4:
1. No change. Because Adrian recommends not changing the function name.
2. Keep Acked-by tag from Adrian.
- Patch5:
1. Add cleanup on trap function as suggested by Ian.
2. Remove Tested-by tag from Adrian because the script is modified.
- Patch6:
1. Add Reviewed-by tag from Ian.
Changes since_v5:
- No code changes.
- Detailed commit message of patch3.
- Add Acked-by and Tested-by tags from Adrian Hunter.
Changes since_v4:
- Simplify check code for record__tracking_system_wide().
- Add perf attr test result to commit message for patch 7.
Changes since_v3:
- Check fall_kernel, all_user, and dummy or exclude_user when determining
whether system wide is required.
Changes since_v2:
- Rename record_tracking.sh to record_sideband.sh in tools/perf/tests/shell.
- Remove "perf evlist: Skip dummy event sample_type check for evlist_config" patch.
- Add opts->all_kernel check in record__config_tracking_events().
- Add perf_event_attr test for record selected CPUs exclude_user.
- Update base-record & system-wide-dummy sample_type attr expected values for test-record-C0.
Changes since v1:
- Add perf_evlist__go_system_wide() via internal/evlist.h instead of
exporting perf_evlist__propagate_maps().
- Use evlist__add_aux_dummy() instead of evlist__add_dummy() in
evlist__findnew_tracking_event().
- Add a parameter in evlist__findnew_tracking_event() to deal with
system_wide inside.
- Add sideband for all CPUs when tracing selected CPUs comments on
the perf record man page.
- Use "sideband events" instead of "tracking events".
- Adjust the patches Sequence.
- Add patch5 to skip dummy event sample_type check for evlist_config.
- Add patch6 to update system-wide-dummy attr values for perf test.
Yang Jihong (6):
perf evlist: Add perf_evlist__go_system_wide() helper
perf evlist: Add evlist__findnew_tracking_event() helper
perf record: Move setting tracking events before
record__init_thread_masks()
perf record: Track sideband events for all CPUs when tracing selected
CPUs
perf test: Add test case for record sideband events
perf test: Add perf_event_attr test for record selected CPUs
exclude_user
Yang Jihong (6):
perf evlist: Add perf_evlist__go_system_wide() helper
perf evlist: Add evlist__findnew_tracking_event() helper
perf record: Move setting tracking events before
record__init_thread_masks()
perf record: Track sideband events for all CPUs when tracing selected
CPUs
perf test: Add test case for record sideband events
perf test: Add perf_event_attr test for record dummy event
tools/lib/perf/evlist.c | 9 +++
tools/lib/perf/include/internal/evlist.h | 2 +
tools/perf/Documentation/perf-record.txt | 3 +
tools/perf/builtin-record.c | 92 +++++++++++++++-------
tools/perf/tests/attr/system-wide-dummy | 14 ++--
tools/perf/tests/attr/test-record-C0 | 4 +-
tools/perf/tests/attr/test-record-dummy-C0 | 55 +++++++++++++
tools/perf/tests/shell/record_sideband.sh | 58 ++++++++++++++
tools/perf/util/evlist.c | 18 +++++
tools/perf/util/evlist.h | 1 +
10 files changed, 221 insertions(+), 35 deletions(-)
create mode 100644 tools/perf/tests/attr/test-record-dummy-C0
create mode 100755 tools/perf/tests/shell/record_sideband.sh
--
2.30.GIT
On 04-Sep-23 8:03 AM, Yang Jihong wrote:
> User space tasks can migrate between CPUs, track sideband events for all
> CPUs.
>
> The specific scenarios are as follows:
>
> CPU0 CPU1
> perf record -C 0 start
> taskA starts to be created and executed
> -> PERF_RECORD_COMM and PERF_RECORD_MMAP
> events only deliver to CPU1
> ......
> |
> migrate to CPU0
> |
> Running on CPU0 <----------/
> ...
>
> perf record -C 0 stop
>
> Now perf samples the PC of taskA. However, perf does not record the
> PERF_RECORD_COMM and PERF_RECORD_COMM events of taskA.
> Therefore, the comm and symbols of taskA cannot be parsed.
>
> The sys_perf_event_open invoked is as follows:
>
> # perf --debug verbose=3 record -e cpu-clock -C 1 true
> <SNIP>
> Opening: cpu-clock
> ------------------------------------------------------------
> perf_event_attr:
> type 1 (PERF_TYPE_SOFTWARE)
> size 136
> config 0 (PERF_COUNT_SW_CPU_CLOCK)
> { sample_period, sample_freq } 4000
> sample_type IP|TID|TIME|CPU|PERIOD|IDENTIFIER
> read_format ID|LOST
> disabled 1
> inherit 1
> freq 1
> sample_id_all 1
> exclude_guest 1
> ------------------------------------------------------------
> sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0x8 = 5
> Opening: dummy:u
> ------------------------------------------------------------
> perf_event_attr:
> type 1 (PERF_TYPE_SOFTWARE)
> size 136
> config 0x9 (PERF_COUNT_SW_DUMMY)
> { sample_period, sample_freq } 1
> sample_type IP|TID|TIME|CPU|IDENTIFIER
> read_format ID|LOST
> inherit 1
> exclude_kernel 1
> exclude_hv 1
> mmap 1
> comm 1
> task 1
> sample_id_all 1
> exclude_guest 1
> mmap2 1
> comm_exec 1
> ksymbol 1
> bpf_event 1
> ------------------------------------------------------------
> sys_perf_event_open: pid -1 cpu 0 group_fd -1 flags 0x8 = 6
> sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0x8 = 7
> sys_perf_event_open: pid -1 cpu 2 group_fd -1 flags 0x8 = 9
> sys_perf_event_open: pid -1 cpu 3 group_fd -1 flags 0x8 = 10
> sys_perf_event_open: pid -1 cpu 4 group_fd -1 flags 0x8 = 11
> sys_perf_event_open: pid -1 cpu 5 group_fd -1 flags 0x8 = 12
> sys_perf_event_open: pid -1 cpu 6 group_fd -1 flags 0x8 = 13
> sys_perf_event_open: pid -1 cpu 7 group_fd -1 flags 0x8 = 14
> <SNIP>
>
> Changes since_v7:
> - The condition for requiring system_wide sideband is changed to
> "as long as a non-dummy event exists" (patch4).
> - Modify the corresponding test case to record only dummy event (patch6).
> - Thanks to tested-by tag from Ravi, but because the solution is modified,
> the tested-by tag of Ravi is not added to this version.
I've re-tested v8 with my simple test.
Tested-by: Ravi Bangoria <ravi.bangoria@amd.com>
Em Tue, Sep 12, 2023 at 02:41:56PM +0530, Ravi Bangoria escreveu:
> On 04-Sep-23 8:03 AM, Yang Jihong wrote:
> > User space tasks can migrate between CPUs, track sideband events for all
> > CPUs.
> >
> > The specific scenarios are as follows:
> >
> > CPU0 CPU1
> > perf record -C 0 start
> > taskA starts to be created and executed
> > -> PERF_RECORD_COMM and PERF_RECORD_MMAP
> > events only deliver to CPU1
> > ......
> > |
> > migrate to CPU0
> > |
> > Running on CPU0 <----------/
> > ...
> >
> > perf record -C 0 stop
> >
> > Now perf samples the PC of taskA. However, perf does not record the
> > PERF_RECORD_COMM and PERF_RECORD_COMM events of taskA.
> > Therefore, the comm and symbols of taskA cannot be parsed.
> >
> > The sys_perf_event_open invoked is as follows:
> >
> > # perf --debug verbose=3 record -e cpu-clock -C 1 true
> > <SNIP>
> > Opening: cpu-clock
> > ------------------------------------------------------------
> > perf_event_attr:
> > type 1 (PERF_TYPE_SOFTWARE)
> > size 136
> > config 0 (PERF_COUNT_SW_CPU_CLOCK)
> > { sample_period, sample_freq } 4000
> > sample_type IP|TID|TIME|CPU|PERIOD|IDENTIFIER
> > read_format ID|LOST
> > disabled 1
> > inherit 1
> > freq 1
> > sample_id_all 1
> > exclude_guest 1
> > ------------------------------------------------------------
> > sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0x8 = 5
> > Opening: dummy:u
> > ------------------------------------------------------------
> > perf_event_attr:
> > type 1 (PERF_TYPE_SOFTWARE)
> > size 136
> > config 0x9 (PERF_COUNT_SW_DUMMY)
> > { sample_period, sample_freq } 1
> > sample_type IP|TID|TIME|CPU|IDENTIFIER
> > read_format ID|LOST
> > inherit 1
> > exclude_kernel 1
> > exclude_hv 1
> > mmap 1
> > comm 1
> > task 1
> > sample_id_all 1
> > exclude_guest 1
> > mmap2 1
> > comm_exec 1
> > ksymbol 1
> > bpf_event 1
> > ------------------------------------------------------------
> > sys_perf_event_open: pid -1 cpu 0 group_fd -1 flags 0x8 = 6
> > sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0x8 = 7
> > sys_perf_event_open: pid -1 cpu 2 group_fd -1 flags 0x8 = 9
> > sys_perf_event_open: pid -1 cpu 3 group_fd -1 flags 0x8 = 10
> > sys_perf_event_open: pid -1 cpu 4 group_fd -1 flags 0x8 = 11
> > sys_perf_event_open: pid -1 cpu 5 group_fd -1 flags 0x8 = 12
> > sys_perf_event_open: pid -1 cpu 6 group_fd -1 flags 0x8 = 13
> > sys_perf_event_open: pid -1 cpu 7 group_fd -1 flags 0x8 = 14
> > <SNIP>
> >
> > Changes since_v7:
> > - The condition for requiring system_wide sideband is changed to
> > "as long as a non-dummy event exists" (patch4).
> > - Modify the corresponding test case to record only dummy event (patch6).
> > - Thanks to tested-by tag from Ravi, but because the solution is modified,
> > the tested-by tag of Ravi is not added to this version.
>
> I've re-tested v8 with my simple test.
>
> Tested-by: Ravi Bangoria <ravi.bangoria@amd.com>
Thanks, applied to the csets that were still sitting in an umpublished
perf-tools-next local branch, soon public.
- Arnaldo
Hello,
On Tue, Sep 12, 2023 at 1:32 PM Arnaldo Carvalho de Melo
<acme@kernel.org> wrote:
>
> Em Tue, Sep 12, 2023 at 02:41:56PM +0530, Ravi Bangoria escreveu:
> > On 04-Sep-23 8:03 AM, Yang Jihong wrote:
> > > User space tasks can migrate between CPUs, track sideband events for all
> > > CPUs.
> > >
> > > The specific scenarios are as follows:
> > >
> > > CPU0 CPU1
> > > perf record -C 0 start
> > > taskA starts to be created and executed
> > > -> PERF_RECORD_COMM and PERF_RECORD_MMAP
> > > events only deliver to CPU1
> > > ......
> > > |
> > > migrate to CPU0
> > > |
> > > Running on CPU0 <----------/
> > > ...
> > >
> > > perf record -C 0 stop
> > >
> > > Now perf samples the PC of taskA. However, perf does not record the
> > > PERF_RECORD_COMM and PERF_RECORD_COMM events of taskA.
> > > Therefore, the comm and symbols of taskA cannot be parsed.
> > >
> > > The sys_perf_event_open invoked is as follows:
> > >
> > > # perf --debug verbose=3 record -e cpu-clock -C 1 true
> > > <SNIP>
> > > Opening: cpu-clock
> > > ------------------------------------------------------------
> > > perf_event_attr:
> > > type 1 (PERF_TYPE_SOFTWARE)
> > > size 136
> > > config 0 (PERF_COUNT_SW_CPU_CLOCK)
> > > { sample_period, sample_freq } 4000
> > > sample_type IP|TID|TIME|CPU|PERIOD|IDENTIFIER
> > > read_format ID|LOST
> > > disabled 1
> > > inherit 1
> > > freq 1
> > > sample_id_all 1
> > > exclude_guest 1
> > > ------------------------------------------------------------
> > > sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0x8 = 5
> > > Opening: dummy:u
> > > ------------------------------------------------------------
> > > perf_event_attr:
> > > type 1 (PERF_TYPE_SOFTWARE)
> > > size 136
> > > config 0x9 (PERF_COUNT_SW_DUMMY)
> > > { sample_period, sample_freq } 1
> > > sample_type IP|TID|TIME|CPU|IDENTIFIER
> > > read_format ID|LOST
> > > inherit 1
> > > exclude_kernel 1
> > > exclude_hv 1
> > > mmap 1
> > > comm 1
> > > task 1
> > > sample_id_all 1
> > > exclude_guest 1
> > > mmap2 1
> > > comm_exec 1
> > > ksymbol 1
> > > bpf_event 1
> > > ------------------------------------------------------------
> > > sys_perf_event_open: pid -1 cpu 0 group_fd -1 flags 0x8 = 6
> > > sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0x8 = 7
> > > sys_perf_event_open: pid -1 cpu 2 group_fd -1 flags 0x8 = 9
> > > sys_perf_event_open: pid -1 cpu 3 group_fd -1 flags 0x8 = 10
> > > sys_perf_event_open: pid -1 cpu 4 group_fd -1 flags 0x8 = 11
> > > sys_perf_event_open: pid -1 cpu 5 group_fd -1 flags 0x8 = 12
> > > sys_perf_event_open: pid -1 cpu 6 group_fd -1 flags 0x8 = 13
> > > sys_perf_event_open: pid -1 cpu 7 group_fd -1 flags 0x8 = 14
> > > <SNIP>
> > >
> > > Changes since_v7:
> > > - The condition for requiring system_wide sideband is changed to
> > > "as long as a non-dummy event exists" (patch4).
> > > - Modify the corresponding test case to record only dummy event (patch6).
> > > - Thanks to tested-by tag from Ravi, but because the solution is modified,
> > > the tested-by tag of Ravi is not added to this version.
> >
> > I've re-tested v8 with my simple test.
> >
> > Tested-by: Ravi Bangoria <ravi.bangoria@amd.com>
>
>
> Thanks, applied to the csets that were still sitting in an umpublished
> perf-tools-next local branch, soon public.
Now I'm seeing a perf test failure on perf-tools-next.
$ sudo ./perf test -v 17
17: Setup struct perf_event_attr :
--- start ---
test child forked, pid 1616372
Using CPUID GenuineIntel-6-8C-1
running './tests/attr/test-record-branch-filter-k'
running './tests/attr/test-record-period'
running './tests/attr/test-record-graph-default'
test limitation '!aarch64'
excluded architecture list ['aarch64']
running './tests/attr/test-record-branch-filter-any'
running './tests/attr/test-record-data'
running './tests/attr/test-stat-detailed-1'
running './tests/attr/test-record-branch-filter-hv'
running './tests/attr/test-record-graph-fp'
test limitation '!aarch64'
excluded architecture list ['aarch64']
running './tests/attr/test-record-basic'
running './tests/attr/test-record-group2'
running './tests/attr/test-stat-detailed-3'
running './tests/attr/test-record-branch-any'
running './tests/attr/test-record-branch-filter-ind_call'
running './tests/attr/test-stat-detailed-2'
running './tests/attr/test-record-group1'
running './tests/attr/test-record-count'
running './tests/attr/test-record-no-samples'
running './tests/attr/test-record-graph-dwarf'
running './tests/attr/test-record-spe-period'
test limitation 'aarch64'
skipped [x86_64] './tests/attr/test-record-spe-period'
running './tests/attr/test-record-graph-fp-aarch64'
test limitation 'aarch64'
skipped [x86_64] './tests/attr/test-record-graph-fp-aarch64'
running './tests/attr/test-record-freq'
running './tests/attr/test-record-pfm-period'
running './tests/attr/test-record-no-buffering'
running './tests/attr/test-record-no-inherit'
running './tests/attr/test-record-branch-filter-any_ret'
running './tests/attr/test-record-raw'
running './tests/attr/test-record-dummy-C0'
expected read_format=4, got 20
FAILED './tests/attr/test-record-dummy-C0' - match failure
test child finished with -1
---- end ----
Setup struct perf_event_attr: FAILED!
Hello,
On 2023/9/16 8:14, Namhyung Kim wrote:
> Hello,
>
> On Tue, Sep 12, 2023 at 1:32 PM Arnaldo Carvalho de Melo
> <acme@kernel.org> wrote:
>>
>> Em Tue, Sep 12, 2023 at 02:41:56PM +0530, Ravi Bangoria escreveu:
>>> On 04-Sep-23 8:03 AM, Yang Jihong wrote:
>>>> User space tasks can migrate between CPUs, track sideband events for all
>>>> CPUs.
>>>>
>>>> The specific scenarios are as follows:
>>>>
>>>> CPU0 CPU1
>>>> perf record -C 0 start
>>>> taskA starts to be created and executed
>>>> -> PERF_RECORD_COMM and PERF_RECORD_MMAP
>>>> events only deliver to CPU1
>>>> ......
>>>> |
>>>> migrate to CPU0
>>>> |
>>>> Running on CPU0 <----------/
>>>> ...
>>>>
>>>> perf record -C 0 stop
>>>>
>>>> Now perf samples the PC of taskA. However, perf does not record the
>>>> PERF_RECORD_COMM and PERF_RECORD_COMM events of taskA.
>>>> Therefore, the comm and symbols of taskA cannot be parsed.
>>>>
>>>> The sys_perf_event_open invoked is as follows:
>>>>
>>>> # perf --debug verbose=3 record -e cpu-clock -C 1 true
>>>> <SNIP>
>>>> Opening: cpu-clock
>>>> ------------------------------------------------------------
>>>> perf_event_attr:
>>>> type 1 (PERF_TYPE_SOFTWARE)
>>>> size 136
>>>> config 0 (PERF_COUNT_SW_CPU_CLOCK)
>>>> { sample_period, sample_freq } 4000
>>>> sample_type IP|TID|TIME|CPU|PERIOD|IDENTIFIER
>>>> read_format ID|LOST
>>>> disabled 1
>>>> inherit 1
>>>> freq 1
>>>> sample_id_all 1
>>>> exclude_guest 1
>>>> ------------------------------------------------------------
>>>> sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0x8 = 5
>>>> Opening: dummy:u
>>>> ------------------------------------------------------------
>>>> perf_event_attr:
>>>> type 1 (PERF_TYPE_SOFTWARE)
>>>> size 136
>>>> config 0x9 (PERF_COUNT_SW_DUMMY)
>>>> { sample_period, sample_freq } 1
>>>> sample_type IP|TID|TIME|CPU|IDENTIFIER
>>>> read_format ID|LOST
>>>> inherit 1
>>>> exclude_kernel 1
>>>> exclude_hv 1
>>>> mmap 1
>>>> comm 1
>>>> task 1
>>>> sample_id_all 1
>>>> exclude_guest 1
>>>> mmap2 1
>>>> comm_exec 1
>>>> ksymbol 1
>>>> bpf_event 1
>>>> ------------------------------------------------------------
>>>> sys_perf_event_open: pid -1 cpu 0 group_fd -1 flags 0x8 = 6
>>>> sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0x8 = 7
>>>> sys_perf_event_open: pid -1 cpu 2 group_fd -1 flags 0x8 = 9
>>>> sys_perf_event_open: pid -1 cpu 3 group_fd -1 flags 0x8 = 10
>>>> sys_perf_event_open: pid -1 cpu 4 group_fd -1 flags 0x8 = 11
>>>> sys_perf_event_open: pid -1 cpu 5 group_fd -1 flags 0x8 = 12
>>>> sys_perf_event_open: pid -1 cpu 6 group_fd -1 flags 0x8 = 13
>>>> sys_perf_event_open: pid -1 cpu 7 group_fd -1 flags 0x8 = 14
>>>> <SNIP>
>>>>
>>>> Changes since_v7:
>>>> - The condition for requiring system_wide sideband is changed to
>>>> "as long as a non-dummy event exists" (patch4).
>>>> - Modify the corresponding test case to record only dummy event (patch6).
>>>> - Thanks to tested-by tag from Ravi, but because the solution is modified,
>>>> the tested-by tag of Ravi is not added to this version.
>>>
>>> I've re-tested v8 with my simple test.
>>>
>>> Tested-by: Ravi Bangoria <ravi.bangoria@amd.com>
>>
>>
>> Thanks, applied to the csets that were still sitting in an umpublished
>> perf-tools-next local branch, soon public.
>
> Now I'm seeing a perf test failure on perf-tools-next.
Uh.. the kernel I was using before didn't support PERF_FORMAT_LOST, so
forget about supporting PERF_FORMAT_LOST. I've updated the kernel and
retested it.
The link to the fixed patch is as follows:
https://lore.kernel.org/all/20230916091641.776031-1-yangjihong1@huawei.com/
Thanks,
Yang
On Sat, Sep 16, 2023 at 2:24 AM Yang Jihong <yangjihong1@huawei.com> wrote:
>
> Hello,
>
> On 2023/9/16 8:14, Namhyung Kim wrote:
> > Hello,
> >
> > On Tue, Sep 12, 2023 at 1:32 PM Arnaldo Carvalho de Melo
> > <acme@kernel.org> wrote:
> >>
> >> Em Tue, Sep 12, 2023 at 02:41:56PM +0530, Ravi Bangoria escreveu:
> >>> On 04-Sep-23 8:03 AM, Yang Jihong wrote:
> >>>> User space tasks can migrate between CPUs, track sideband events for all
> >>>> CPUs.
> >>>>
> >>>> The specific scenarios are as follows:
> >>>>
> >>>> CPU0 CPU1
> >>>> perf record -C 0 start
> >>>> taskA starts to be created and executed
> >>>> -> PERF_RECORD_COMM and PERF_RECORD_MMAP
> >>>> events only deliver to CPU1
> >>>> ......
> >>>> |
> >>>> migrate to CPU0
> >>>> |
> >>>> Running on CPU0 <----------/
> >>>> ...
> >>>>
> >>>> perf record -C 0 stop
> >>>>
> >>>> Now perf samples the PC of taskA. However, perf does not record the
> >>>> PERF_RECORD_COMM and PERF_RECORD_COMM events of taskA.
> >>>> Therefore, the comm and symbols of taskA cannot be parsed.
> >>>>
> >>>> The sys_perf_event_open invoked is as follows:
> >>>>
> >>>> # perf --debug verbose=3 record -e cpu-clock -C 1 true
> >>>> <SNIP>
> >>>> Opening: cpu-clock
> >>>> ------------------------------------------------------------
> >>>> perf_event_attr:
> >>>> type 1 (PERF_TYPE_SOFTWARE)
> >>>> size 136
> >>>> config 0 (PERF_COUNT_SW_CPU_CLOCK)
> >>>> { sample_period, sample_freq } 4000
> >>>> sample_type IP|TID|TIME|CPU|PERIOD|IDENTIFIER
> >>>> read_format ID|LOST
> >>>> disabled 1
> >>>> inherit 1
> >>>> freq 1
> >>>> sample_id_all 1
> >>>> exclude_guest 1
> >>>> ------------------------------------------------------------
> >>>> sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0x8 = 5
> >>>> Opening: dummy:u
> >>>> ------------------------------------------------------------
> >>>> perf_event_attr:
> >>>> type 1 (PERF_TYPE_SOFTWARE)
> >>>> size 136
> >>>> config 0x9 (PERF_COUNT_SW_DUMMY)
> >>>> { sample_period, sample_freq } 1
> >>>> sample_type IP|TID|TIME|CPU|IDENTIFIER
> >>>> read_format ID|LOST
> >>>> inherit 1
> >>>> exclude_kernel 1
> >>>> exclude_hv 1
> >>>> mmap 1
> >>>> comm 1
> >>>> task 1
> >>>> sample_id_all 1
> >>>> exclude_guest 1
> >>>> mmap2 1
> >>>> comm_exec 1
> >>>> ksymbol 1
> >>>> bpf_event 1
> >>>> ------------------------------------------------------------
> >>>> sys_perf_event_open: pid -1 cpu 0 group_fd -1 flags 0x8 = 6
> >>>> sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0x8 = 7
> >>>> sys_perf_event_open: pid -1 cpu 2 group_fd -1 flags 0x8 = 9
> >>>> sys_perf_event_open: pid -1 cpu 3 group_fd -1 flags 0x8 = 10
> >>>> sys_perf_event_open: pid -1 cpu 4 group_fd -1 flags 0x8 = 11
> >>>> sys_perf_event_open: pid -1 cpu 5 group_fd -1 flags 0x8 = 12
> >>>> sys_perf_event_open: pid -1 cpu 6 group_fd -1 flags 0x8 = 13
> >>>> sys_perf_event_open: pid -1 cpu 7 group_fd -1 flags 0x8 = 14
> >>>> <SNIP>
> >>>>
> >>>> Changes since_v7:
> >>>> - The condition for requiring system_wide sideband is changed to
> >>>> "as long as a non-dummy event exists" (patch4).
> >>>> - Modify the corresponding test case to record only dummy event (patch6).
> >>>> - Thanks to tested-by tag from Ravi, but because the solution is modified,
> >>>> the tested-by tag of Ravi is not added to this version.
> >>>
> >>> I've re-tested v8 with my simple test.
> >>>
> >>> Tested-by: Ravi Bangoria <ravi.bangoria@amd.com>
> >>
> >>
> >> Thanks, applied to the csets that were still sitting in an umpublished
> >> perf-tools-next local branch, soon public.
> >
> > Now I'm seeing a perf test failure on perf-tools-next.
>
> Uh.. the kernel I was using before didn't support PERF_FORMAT_LOST, so
> forget about supporting PERF_FORMAT_LOST. I've updated the kernel and
> retested it.
>
> The link to the fixed patch is as follows:
> https://lore.kernel.org/all/20230916091641.776031-1-yangjihong1@huawei.com/
Thank you for the quick fix!
Namhyung
Em Mon, Sep 04, 2023 at 02:33:34AM +0000, Yang Jihong escreveu:
> User space tasks can migrate between CPUs, track sideband events for all
> CPUs.
>
> The specific scenarios are as follows:
>
> CPU0 CPU1
> perf record -C 0 start
> taskA starts to be created and executed
> -> PERF_RECORD_COMM and PERF_RECORD_MMAP
> events only deliver to CPU1
> ......
> |
> migrate to CPU0
> |
> Running on CPU0 <----------/
> ...
>
> perf record -C 0 stop
>
> Now perf samples the PC of taskA. However, perf does not record the
> PERF_RECORD_COMM and PERF_RECORD_COMM events of taskA.
> Therefore, the comm and symbols of taskA cannot be parsed.
>
> The sys_perf_event_open invoked is as follows:
>
> # perf --debug verbose=3 record -e cpu-clock -C 1 true
> <SNIP>
> Opening: cpu-clock
> ------------------------------------------------------------
> perf_event_attr:
> type 1 (PERF_TYPE_SOFTWARE)
> size 136
> config 0 (PERF_COUNT_SW_CPU_CLOCK)
> { sample_period, sample_freq } 4000
> sample_type IP|TID|TIME|CPU|PERIOD|IDENTIFIER
> read_format ID|LOST
> disabled 1
> inherit 1
> freq 1
> sample_id_all 1
> exclude_guest 1
> ------------------------------------------------------------
> sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0x8 = 5
> Opening: dummy:u
> ------------------------------------------------------------
> perf_event_attr:
> type 1 (PERF_TYPE_SOFTWARE)
> size 136
> config 0x9 (PERF_COUNT_SW_DUMMY)
> { sample_period, sample_freq } 1
> sample_type IP|TID|TIME|CPU|IDENTIFIER
> read_format ID|LOST
> inherit 1
> exclude_kernel 1
> exclude_hv 1
> mmap 1
> comm 1
> task 1
> sample_id_all 1
> exclude_guest 1
> mmap2 1
> comm_exec 1
> ksymbol 1
> bpf_event 1
> ------------------------------------------------------------
> sys_perf_event_open: pid -1 cpu 0 group_fd -1 flags 0x8 = 6
> sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0x8 = 7
> sys_perf_event_open: pid -1 cpu 2 group_fd -1 flags 0x8 = 9
> sys_perf_event_open: pid -1 cpu 3 group_fd -1 flags 0x8 = 10
> sys_perf_event_open: pid -1 cpu 4 group_fd -1 flags 0x8 = 11
> sys_perf_event_open: pid -1 cpu 5 group_fd -1 flags 0x8 = 12
> sys_perf_event_open: pid -1 cpu 6 group_fd -1 flags 0x8 = 13
> sys_perf_event_open: pid -1 cpu 7 group_fd -1 flags 0x8 = 14
> <SNIP>
>
> Changes since_v7:
> - The condition for requiring system_wide sideband is changed to
> "as long as a non-dummy event exists" (patch4).
> - Modify the corresponding test case to record only dummy event (patch6).
> - Thanks to tested-by tag from Ravi, but because the solution is modified,
> the tested-by tag of Ravi is not added to this version.
>
> Changes since_v6:
> - Patch1:
> 1. No change.
> 2. Keep Acked-by tag from Adrian.
> - Patch2:
> 1. Update commit message as suggested by Ian.
> 2. Keep Acked-by tag from Adrian because code is not modified.
> - Patch3:
> 1. Update comment as suggested by Ian.
> 2. Merge original patch5 ("perf test: Update base-record & system-wide-dummy attr") as suggested by Ian.
> 3. Only merge commit, keep Acked-by tag from Adrian.
> - Patch4:
> 1. No change. Because Adrian recommends not changing the function name.
> 2. Keep Acked-by tag from Adrian.
> - Patch5:
> 1. Add cleanup on trap function as suggested by Ian.
> 2. Remove Tested-by tag from Adrian because the script is modified.
> - Patch6:
> 1. Add Reviewed-by tag from Ian.
I'm in doubt about these Acked-by/Reviewed-by tags, do they still stand? They are
not in the latest series, can you please check?
- Arnaldo
> Changes since_v5:
> - No code changes.
> - Detailed commit message of patch3.
> - Add Acked-by and Tested-by tags from Adrian Hunter.
>
> Changes since_v4:
> - Simplify check code for record__tracking_system_wide().
> - Add perf attr test result to commit message for patch 7.
>
> Changes since_v3:
> - Check fall_kernel, all_user, and dummy or exclude_user when determining
> whether system wide is required.
>
> Changes since_v2:
> - Rename record_tracking.sh to record_sideband.sh in tools/perf/tests/shell.
> - Remove "perf evlist: Skip dummy event sample_type check for evlist_config" patch.
> - Add opts->all_kernel check in record__config_tracking_events().
> - Add perf_event_attr test for record selected CPUs exclude_user.
> - Update base-record & system-wide-dummy sample_type attr expected values for test-record-C0.
>
> Changes since v1:
> - Add perf_evlist__go_system_wide() via internal/evlist.h instead of
> exporting perf_evlist__propagate_maps().
> - Use evlist__add_aux_dummy() instead of evlist__add_dummy() in
> evlist__findnew_tracking_event().
> - Add a parameter in evlist__findnew_tracking_event() to deal with
> system_wide inside.
> - Add sideband for all CPUs when tracing selected CPUs comments on
> the perf record man page.
> - Use "sideband events" instead of "tracking events".
> - Adjust the patches Sequence.
> - Add patch5 to skip dummy event sample_type check for evlist_config.
> - Add patch6 to update system-wide-dummy attr values for perf test.
>
> Yang Jihong (6):
> perf evlist: Add perf_evlist__go_system_wide() helper
> perf evlist: Add evlist__findnew_tracking_event() helper
> perf record: Move setting tracking events before
> record__init_thread_masks()
> perf record: Track sideband events for all CPUs when tracing selected
> CPUs
> perf test: Add test case for record sideband events
> perf test: Add perf_event_attr test for record selected CPUs
> exclude_user
>
> Yang Jihong (6):
> perf evlist: Add perf_evlist__go_system_wide() helper
> perf evlist: Add evlist__findnew_tracking_event() helper
> perf record: Move setting tracking events before
> record__init_thread_masks()
> perf record: Track sideband events for all CPUs when tracing selected
> CPUs
> perf test: Add test case for record sideband events
> perf test: Add perf_event_attr test for record dummy event
>
> tools/lib/perf/evlist.c | 9 +++
> tools/lib/perf/include/internal/evlist.h | 2 +
> tools/perf/Documentation/perf-record.txt | 3 +
> tools/perf/builtin-record.c | 92 +++++++++++++++-------
> tools/perf/tests/attr/system-wide-dummy | 14 ++--
> tools/perf/tests/attr/test-record-C0 | 4 +-
> tools/perf/tests/attr/test-record-dummy-C0 | 55 +++++++++++++
> tools/perf/tests/shell/record_sideband.sh | 58 ++++++++++++++
> tools/perf/util/evlist.c | 18 +++++
> tools/perf/util/evlist.h | 1 +
> 10 files changed, 221 insertions(+), 35 deletions(-)
> create mode 100644 tools/perf/tests/attr/test-record-dummy-C0
> create mode 100755 tools/perf/tests/shell/record_sideband.sh
>
> --
> 2.30.GIT
>
--
- Arnaldo
Hello,
On 2023/9/7 0:08, Arnaldo Carvalho de Melo wrote:
> Em Mon, Sep 04, 2023 at 02:33:34AM +0000, Yang Jihong escreveu:
>> User space tasks can migrate between CPUs, track sideband events for all
>> CPUs.
>>
>> The specific scenarios are as follows:
>>
>> CPU0 CPU1
>> perf record -C 0 start
>> taskA starts to be created and executed
>> -> PERF_RECORD_COMM and PERF_RECORD_MMAP
>> events only deliver to CPU1
>> ......
>> |
>> migrate to CPU0
>> |
>> Running on CPU0 <----------/
>> ...
>>
>> perf record -C 0 stop
>>
>> Now perf samples the PC of taskA. However, perf does not record the
>> PERF_RECORD_COMM and PERF_RECORD_COMM events of taskA.
>> Therefore, the comm and symbols of taskA cannot be parsed.
>>
>> The sys_perf_event_open invoked is as follows:
>>
>> # perf --debug verbose=3 record -e cpu-clock -C 1 true
>> <SNIP>
>> Opening: cpu-clock
>> ------------------------------------------------------------
>> perf_event_attr:
>> type 1 (PERF_TYPE_SOFTWARE)
>> size 136
>> config 0 (PERF_COUNT_SW_CPU_CLOCK)
>> { sample_period, sample_freq } 4000
>> sample_type IP|TID|TIME|CPU|PERIOD|IDENTIFIER
>> read_format ID|LOST
>> disabled 1
>> inherit 1
>> freq 1
>> sample_id_all 1
>> exclude_guest 1
>> ------------------------------------------------------------
>> sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0x8 = 5
>> Opening: dummy:u
>> ------------------------------------------------------------
>> perf_event_attr:
>> type 1 (PERF_TYPE_SOFTWARE)
>> size 136
>> config 0x9 (PERF_COUNT_SW_DUMMY)
>> { sample_period, sample_freq } 1
>> sample_type IP|TID|TIME|CPU|IDENTIFIER
>> read_format ID|LOST
>> inherit 1
>> exclude_kernel 1
>> exclude_hv 1
>> mmap 1
>> comm 1
>> task 1
>> sample_id_all 1
>> exclude_guest 1
>> mmap2 1
>> comm_exec 1
>> ksymbol 1
>> bpf_event 1
>> ------------------------------------------------------------
>> sys_perf_event_open: pid -1 cpu 0 group_fd -1 flags 0x8 = 6
>> sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0x8 = 7
>> sys_perf_event_open: pid -1 cpu 2 group_fd -1 flags 0x8 = 9
>> sys_perf_event_open: pid -1 cpu 3 group_fd -1 flags 0x8 = 10
>> sys_perf_event_open: pid -1 cpu 4 group_fd -1 flags 0x8 = 11
>> sys_perf_event_open: pid -1 cpu 5 group_fd -1 flags 0x8 = 12
>> sys_perf_event_open: pid -1 cpu 6 group_fd -1 flags 0x8 = 13
>> sys_perf_event_open: pid -1 cpu 7 group_fd -1 flags 0x8 = 14
>> <SNIP>
>>
>> Changes since_v7:
>> - The condition for requiring system_wide sideband is changed to
>> "as long as a non-dummy event exists" (patch4).
>> - Modify the corresponding test case to record only dummy event (patch6).
>> - Thanks to tested-by tag from Ravi, but because the solution is modified,
>> the tested-by tag of Ravi is not added to this version.
>>
>> Changes since_v6:
>> - Patch1:
>> 1. No change.
>> 2. Keep Acked-by tag from Adrian.
>> - Patch2:
>> 1. Update commit message as suggested by Ian.
>> 2. Keep Acked-by tag from Adrian because code is not modified.
>> - Patch3:
>> 1. Update comment as suggested by Ian.
>> 2. Merge original patch5 ("perf test: Update base-record & system-wide-dummy attr") as suggested by Ian.
>> 3. Only merge commit, keep Acked-by tag from Adrian.
>> - Patch4:
>> 1. No change. Because Adrian recommends not changing the function name.
>> 2. Keep Acked-by tag from Adrian.
>> - Patch5:
>> 1. Add cleanup on trap function as suggested by Ian.
>> 2. Remove Tested-by tag from Adrian because the script is modified.
>> - Patch6:
>> 1. Add Reviewed-by tag from Ian.
>
> I'm in doubt about these Acked-by/Reviewed-by tags, do they still stand? They are
> not in the latest series, can you please check?
Uh, uh. Because several reviewers have different opinions on the
solution, I modified it several times.
Now v8 patchset is different from the previous versions.
I only keep the Acked-by/Reviewed-by tags of patches that are not
modified. For patches that have modified the code, I remove the tags.
Therefore, please refer to the v8 series for Acked-by/Reviewed-by tags.
This version needs to be confirmed by reviewers.
Thanks,
Yang
© 2016 - 2025 Red Hat, Inc.